def testEventDataset(self):
# baisc checks
self.failUnlessRaises(DatasetError, EventDataset)
# simple data
samples = N.arange(240).reshape(10, 2, 3, 4)
# copy constructor does not work on non-2D data
self.failUnlessRaises(DatasetError, EventDataset, samples=samples)
# try case without extra features
evs = [Event(onset=2, duration=2, label=1, chunk=2),
Event(onset=5, duration=1, label=2, chunk=2),
Event(onset=7, duration=2, label=3, chunk=4)]
ds = EventDataset(samples=samples, events=evs)
self.failUnless(ds.nfeatures == 48)
self.failUnless(ds.nsamples == 3)
self.failUnless((ds.labels == [1,2,3]).all())
self.failUnless((ds.chunks == [2,2,4]).all())
mr = ds.mapReverse(N.arange(48))
self.failUnless((mr == N.arange(48).reshape(2,2,3,4)).all())
# try case with extra features
evs = [Event(onset=2, duration=2, label=1, features=[2,3]),
Event(onset=5, duration=2, label=1, features=[4,5]),
Event(onset=7, duration=2, label=1, features=[6,7]),]
ds = EventDataset(samples=samples, events=evs)
# we have 2 additional features
self.failUnless(ds.nfeatures == 50)
self.failUnless(ds.nsamples == 3)
self.failUnless((ds.labels == [1,1,1]).all())
self.failUnless((ds.chunks == [0,1,2]).all())
# now for the long awaited -- map back into two distinct
# feature spaces
mr = ds.mapReverse(N.arange(50))
# we get two sets of feature spaces (samples and extra features)
self.failUnless(len(mr) == 2)
msamples, mxfeat = mr
# the sample side should be identical to the case without extra features
self.failUnless((msamples == N.arange(48).reshape(2,2,3,4)).all())
# the extra features should be flat
self.failUnless((mxfeat == (48,49)).all())
# now take a look at
orig = ds.O
self.failUnless(len(mr) == 2)
osamples, oextra = orig
self.failUnless((oextra == [[2,3],[4,5],[6,7]]).all())
self.failUnless(osamples.shape == samples.shape)
# check that all samples not covered by an event are zero
filt = N.array([True,True,False,False,True,False,False,False,False,True])
self.failUnless(N.sum(osamples[filt]) == 0)
self.failUnless((osamples[N.negative(filt)] > 0).all())
def __init__(self, samples=None, events=None, mask=None, evconv=False,
storeoffset=False, tr=None, enforce_dim=4, **kwargs):
"""
:Paramaters:
mask: str | NiftiImage | ndarray
Filename of a NIfTI image or a `NiftiImage` instance or an ndarray
of appropriate shape.
evconv: bool
Convert event definitions using `onset` and `duration` in some
temporal unit into #sample notation.
storeoffset: Bool
Whether to store temproal offset information when converting
Events into descrete time. Only considered when evconv == True.
tr: float
Temporal distance of two adjacent NIfTI volumes. This can be used
to override the corresponding value in the NIfTI header.
enforce_dim : int or None
If not None, it is the dimensionality of the data to be enforced,
commonly 4D for the data, and 3D for the mask in case of fMRI.
"""
# check if we are in copy constructor mode
if events is None:
EventDataset.__init__(self, samples=samples, events=events,
mask=mask, **kwargs)
return
nifti = getNiftiFromAnySource(samples, ensure=True,
enforce_dim=enforce_dim)
# no copying
samples = nifti.data
# do not put the whole NiftiImage in the dict as this will most
# likely be deepcopy'ed at some point and ensuring data integrity
# of the complex Python-C-Swig hybrid might be a tricky task.
# Only storing the header dict should achieve the same and is more
# memory efficient and even simpler
dsattr = {'niftihdr': nifti.header}
# determine TR, take from NIfTI header by default
dt = nifti.rtime
# override if necessary
if not tr is None:
dt = tr
# NiftiDataset uses a DescreteMetric with cartesian
# distance and element size from the NIfTI header
# 'voxdim' is (x,y,z) while 'samples' are (t,z,y,x)
elementsize = [dt] + [i for i in reversed(nifti.voxdim)]
# XXX metric might be inappropriate if boxcar has length 1
# might move metric setup after baseclass init and check what has
# really happened
metric = DescreteMetric(elementsize=elementsize,
distance_function=cartesianDistance)
# convert EVs if necessary -- not altering original
if evconv:
if dt == 0:
raise ValueError, "'dt' cannot be zero when converting Events"
events = [ev.asDescreteTime(dt, storeoffset) for ev in events]
else:
# do not touch the original
events = deepcopy(events)
# forcefully convert onset and duration into integers, as expected
# by the baseclass
for ev in events:
oldonset = ev['onset']
oldduration = ev['duration']
ev['onset'] = int(ev['onset'])
ev['duration'] = int(ev['duration'])
if not oldonset == ev['onset'] \
or not oldduration == ev['duration']:
warning("Loosing information during automatic integer "
"conversion of EVs. Consider an explicit conversion"
" by setting `evconv` in ERNiftiDataset().")
# pull mask array from NIfTI (if present)
if mask is None:
pass
elif isinstance(mask, N.ndarray):
# plain array can be passed on to base class
pass
else:
mask_nim = getNiftiFromAnySource(mask)
if not mask_nim is None:
mask = getNiftiData(mask_nim)
else:
raise ValueError, "Cannot load mask from '%s'" % mask
# finally init baseclass
EventDataset.__init__(self, samples=samples, events=events,
mask=mask, dametric=metric, dsattr=dsattr,
**kwargs)
